AI悖论：优化如何可能削弱助人职业中的意义性工作

Helping professions are adopting artificial intelligence (AI) to automate documentation, monitoring, and coordination on the premise that offloading routine tasks frees workers for meaningful care. That premise assumes managers know which tasks are routine and which meaningful. We argue instead for a task-contingent AI paradox: meaning in helping work concentrates in identity-central tasks that are increasingly automatable. Study 1, a two-wave critical-incident panel of 307 helping-profession workers (144 longitudinal cases one year later), tested six pre-specified predictions that Time 1 human-coded meaning dimensions forecast Time 2 felt meaningfulness, exhaustion, or turnover intentions; estimates were near zero, and equivalence at the |β| = .30 sensitivity band held for three of five. One exploratory association survived multiplicity screening: narratives grounding meaning in competence were associated with higher later turnover intentions and exhaustion (β = .29 and .30, adjusting for six survey measures). Automated coding did not reach the human-agreement standard; humans coded all predictors. Study 2 coded 7,695 nursing-forum posts captured 2016–2026 for meaning events. Nurses named AI in seven events; about one post in twenty per period instead named an optimization or implementation object (ratios, metrics, dashboards, documentation load), and the facet at issue was control over means. Study 3 found that share unchanged across the decade; the one strain marker that rose, lacking control, is confounded with a site template change. No study measured individual AI exposure. Against the automation frontier, the results locate the risk of optimization in the tasks workers name as meaningful, not in AI as such.